Home | History | Annotate | Line # | Download | only in kern
kern_time.c revision 1.179.10.3
      1 /*	$NetBSD: kern_time.c,v 1.179.10.3 2016/03/19 11:30:31 skrll Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2000, 2004, 2005, 2007, 2008, 2009 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Christopher G. Demetriou, and by Andrew Doran.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 1982, 1986, 1989, 1993
     34  *	The Regents of the University of California.  All rights reserved.
     35  *
     36  * Redistribution and use in source and binary forms, with or without
     37  * modification, are permitted provided that the following conditions
     38  * are met:
     39  * 1. Redistributions of source code must retain the above copyright
     40  *    notice, this list of conditions and the following disclaimer.
     41  * 2. Redistributions in binary form must reproduce the above copyright
     42  *    notice, this list of conditions and the following disclaimer in the
     43  *    documentation and/or other materials provided with the distribution.
     44  * 3. Neither the name of the University nor the names of its contributors
     45  *    may be used to endorse or promote products derived from this software
     46  *    without specific prior written permission.
     47  *
     48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     58  * SUCH DAMAGE.
     59  *
     60  *	@(#)kern_time.c	8.4 (Berkeley) 5/26/95
     61  */
     62 
     63 #include <sys/cdefs.h>
     64 __KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.179.10.3 2016/03/19 11:30:31 skrll Exp $");
     65 
     66 #include <sys/param.h>
     67 #include <sys/resourcevar.h>
     68 #include <sys/kernel.h>
     69 #include <sys/systm.h>
     70 #include <sys/proc.h>
     71 #include <sys/vnode.h>
     72 #include <sys/signalvar.h>
     73 #include <sys/syslog.h>
     74 #include <sys/timetc.h>
     75 #include <sys/timex.h>
     76 #include <sys/kauth.h>
     77 #include <sys/mount.h>
     78 #include <sys/syscallargs.h>
     79 #include <sys/cpu.h>
     80 
     81 static void	timer_intr(void *);
     82 static void	itimerfire(struct ptimer *);
     83 static void	itimerfree(struct ptimers *, int);
     84 
     85 kmutex_t	timer_lock;
     86 
     87 static void	*timer_sih;
     88 static TAILQ_HEAD(, ptimer) timer_queue;
     89 
     90 struct pool ptimer_pool, ptimers_pool;
     91 
     92 #define	CLOCK_VIRTUAL_P(clockid)	\
     93 	((clockid) == CLOCK_VIRTUAL || (clockid) == CLOCK_PROF)
     94 
     95 CTASSERT(ITIMER_REAL == CLOCK_REALTIME);
     96 CTASSERT(ITIMER_VIRTUAL == CLOCK_VIRTUAL);
     97 CTASSERT(ITIMER_PROF == CLOCK_PROF);
     98 CTASSERT(ITIMER_MONOTONIC == CLOCK_MONOTONIC);
     99 
    100 /*
    101  * Initialize timekeeping.
    102  */
    103 void
    104 time_init(void)
    105 {
    106 
    107 	pool_init(&ptimer_pool, sizeof(struct ptimer), 0, 0, 0, "ptimerpl",
    108 	    &pool_allocator_nointr, IPL_NONE);
    109 	pool_init(&ptimers_pool, sizeof(struct ptimers), 0, 0, 0, "ptimerspl",
    110 	    &pool_allocator_nointr, IPL_NONE);
    111 }
    112 
    113 void
    114 time_init2(void)
    115 {
    116 
    117 	TAILQ_INIT(&timer_queue);
    118 	mutex_init(&timer_lock, MUTEX_DEFAULT, IPL_SCHED);
    119 	timer_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
    120 	    timer_intr, NULL);
    121 }
    122 
    123 /* Time of day and interval timer support.
    124  *
    125  * These routines provide the kernel entry points to get and set
    126  * the time-of-day and per-process interval timers.  Subroutines
    127  * here provide support for adding and subtracting timeval structures
    128  * and decrementing interval timers, optionally reloading the interval
    129  * timers when they expire.
    130  */
    131 
    132 /* This function is used by clock_settime and settimeofday */
    133 static int
    134 settime1(struct proc *p, const struct timespec *ts, bool check_kauth)
    135 {
    136 	struct timespec delta, now;
    137 	int s;
    138 
    139 	/* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */
    140 	s = splclock();
    141 	nanotime(&now);
    142 	timespecsub(ts, &now, &delta);
    143 
    144 	if (check_kauth && kauth_authorize_system(kauth_cred_get(),
    145 	    KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_SYSTEM, __UNCONST(ts),
    146 	    &delta, KAUTH_ARG(check_kauth ? false : true)) != 0) {
    147 		splx(s);
    148 		return (EPERM);
    149 	}
    150 
    151 #ifdef notyet
    152 	if ((delta.tv_sec < 86400) && securelevel > 0) { /* XXX elad - notyet */
    153 		splx(s);
    154 		return (EPERM);
    155 	}
    156 #endif
    157 
    158 	tc_setclock(ts);
    159 
    160 	timespecadd(&boottime, &delta, &boottime);
    161 
    162 	resettodr();
    163 	splx(s);
    164 
    165 	return (0);
    166 }
    167 
    168 int
    169 settime(struct proc *p, struct timespec *ts)
    170 {
    171 	return (settime1(p, ts, true));
    172 }
    173 
    174 /* ARGSUSED */
    175 int
    176 sys___clock_gettime50(struct lwp *l,
    177     const struct sys___clock_gettime50_args *uap, register_t *retval)
    178 {
    179 	/* {
    180 		syscallarg(clockid_t) clock_id;
    181 		syscallarg(struct timespec *) tp;
    182 	} */
    183 	int error;
    184 	struct timespec ats;
    185 
    186 	error = clock_gettime1(SCARG(uap, clock_id), &ats);
    187 	if (error != 0)
    188 		return error;
    189 
    190 	return copyout(&ats, SCARG(uap, tp), sizeof(ats));
    191 }
    192 
    193 /* ARGSUSED */
    194 int
    195 sys___clock_settime50(struct lwp *l,
    196     const struct sys___clock_settime50_args *uap, register_t *retval)
    197 {
    198 	/* {
    199 		syscallarg(clockid_t) clock_id;
    200 		syscallarg(const struct timespec *) tp;
    201 	} */
    202 	int error;
    203 	struct timespec ats;
    204 
    205 	if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
    206 		return error;
    207 
    208 	return clock_settime1(l->l_proc, SCARG(uap, clock_id), &ats, true);
    209 }
    210 
    211 
    212 int
    213 clock_settime1(struct proc *p, clockid_t clock_id, const struct timespec *tp,
    214     bool check_kauth)
    215 {
    216 	int error;
    217 
    218 	switch (clock_id) {
    219 	case CLOCK_REALTIME:
    220 		if ((error = settime1(p, tp, check_kauth)) != 0)
    221 			return (error);
    222 		break;
    223 	case CLOCK_MONOTONIC:
    224 		return (EINVAL);	/* read-only clock */
    225 	default:
    226 		return (EINVAL);
    227 	}
    228 
    229 	return 0;
    230 }
    231 
    232 int
    233 sys___clock_getres50(struct lwp *l, const struct sys___clock_getres50_args *uap,
    234     register_t *retval)
    235 {
    236 	/* {
    237 		syscallarg(clockid_t) clock_id;
    238 		syscallarg(struct timespec *) tp;
    239 	} */
    240 	struct timespec ts;
    241 	int error;
    242 
    243 	if ((error = clock_getres1(SCARG(uap, clock_id), &ts)) != 0)
    244 		return error;
    245 
    246 	if (SCARG(uap, tp))
    247 		error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
    248 
    249 	return error;
    250 }
    251 
    252 int
    253 clock_getres1(clockid_t clock_id, struct timespec *ts)
    254 {
    255 
    256 	switch (clock_id) {
    257 	case CLOCK_REALTIME:
    258 	case CLOCK_MONOTONIC:
    259 		ts->tv_sec = 0;
    260 		if (tc_getfrequency() > 1000000000)
    261 			ts->tv_nsec = 1;
    262 		else
    263 			ts->tv_nsec = 1000000000 / tc_getfrequency();
    264 		break;
    265 	default:
    266 		return EINVAL;
    267 	}
    268 
    269 	return 0;
    270 }
    271 
    272 /* ARGSUSED */
    273 int
    274 sys___nanosleep50(struct lwp *l, const struct sys___nanosleep50_args *uap,
    275     register_t *retval)
    276 {
    277 	/* {
    278 		syscallarg(struct timespec *) rqtp;
    279 		syscallarg(struct timespec *) rmtp;
    280 	} */
    281 	struct timespec rmt, rqt;
    282 	int error, error1;
    283 
    284 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
    285 	if (error)
    286 		return (error);
    287 
    288 	error = nanosleep1(l, CLOCK_MONOTONIC, 0, &rqt,
    289 	    SCARG(uap, rmtp) ? &rmt : NULL);
    290 	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
    291 		return error;
    292 
    293 	error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt));
    294 	return error1 ? error1 : error;
    295 }
    296 
    297 /* ARGSUSED */
    298 int
    299 sys_clock_nanosleep(struct lwp *l, const struct sys_clock_nanosleep_args *uap,
    300     register_t *retval)
    301 {
    302 	/* {
    303 		syscallarg(clockid_t) clock_id;
    304 		syscallarg(int) flags;
    305 		syscallarg(struct timespec *) rqtp;
    306 		syscallarg(struct timespec *) rmtp;
    307 	} */
    308 	struct timespec rmt, rqt;
    309 	int error, error1;
    310 
    311 	error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec));
    312 	if (error)
    313 		goto out;
    314 
    315 	error = nanosleep1(l, SCARG(uap, clock_id), SCARG(uap, flags), &rqt,
    316 	    SCARG(uap, rmtp) ? &rmt : NULL);
    317 	if (SCARG(uap, rmtp) == NULL || (error != 0 && error != EINTR))
    318 		goto out;
    319 
    320 	if ((error1 = copyout(&rmt, SCARG(uap, rmtp), sizeof(rmt))) != 0)
    321 		error = error1;
    322 out:
    323 	*retval = error;
    324 	return 0;
    325 }
    326 
    327 int
    328 nanosleep1(struct lwp *l, clockid_t clock_id, int flags, struct timespec *rqt,
    329     struct timespec *rmt)
    330 {
    331 	struct timespec rmtstart;
    332 	int error, timo;
    333 
    334 	if ((error = ts2timo(clock_id, flags, rqt, &timo, &rmtstart)) != 0) {
    335 		if (error == ETIMEDOUT) {
    336 			error = 0;
    337 			if (rmt != NULL)
    338 				rmt->tv_sec = rmt->tv_nsec = 0;
    339 		}
    340 		return error;
    341 	}
    342 
    343 	/*
    344 	 * Avoid inadvertently sleeping forever
    345 	 */
    346 	if (timo == 0)
    347 		timo = 1;
    348 again:
    349 	error = kpause("nanoslp", true, timo, NULL);
    350 	if (rmt != NULL || error == 0) {
    351 		struct timespec rmtend;
    352 		struct timespec t0;
    353 		struct timespec *t;
    354 
    355 		(void)clock_gettime1(clock_id, &rmtend);
    356 		t = (rmt != NULL) ? rmt : &t0;
    357 		if (flags & TIMER_ABSTIME) {
    358 			timespecsub(rqt, &rmtend, t);
    359 		} else {
    360 			timespecsub(&rmtend, &rmtstart, t);
    361 			timespecsub(rqt, t, t);
    362 		}
    363 		if (t->tv_sec < 0)
    364 			timespecclear(t);
    365 		if (error == 0) {
    366 			timo = tstohz(t);
    367 			if (timo > 0)
    368 				goto again;
    369 		}
    370 	}
    371 
    372 	if (error == ERESTART)
    373 		error = EINTR;
    374 	if (error == EWOULDBLOCK)
    375 		error = 0;
    376 
    377 	return error;
    378 }
    379 
    380 /* ARGSUSED */
    381 int
    382 sys___gettimeofday50(struct lwp *l, const struct sys___gettimeofday50_args *uap,
    383     register_t *retval)
    384 {
    385 	/* {
    386 		syscallarg(struct timeval *) tp;
    387 		syscallarg(void *) tzp;		really "struct timezone *";
    388 	} */
    389 	struct timeval atv;
    390 	int error = 0;
    391 	struct timezone tzfake;
    392 
    393 	if (SCARG(uap, tp)) {
    394 		microtime(&atv);
    395 		error = copyout(&atv, SCARG(uap, tp), sizeof(atv));
    396 		if (error)
    397 			return (error);
    398 	}
    399 	if (SCARG(uap, tzp)) {
    400 		/*
    401 		 * NetBSD has no kernel notion of time zone, so we just
    402 		 * fake up a timezone struct and return it if demanded.
    403 		 */
    404 		tzfake.tz_minuteswest = 0;
    405 		tzfake.tz_dsttime = 0;
    406 		error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake));
    407 	}
    408 	return (error);
    409 }
    410 
    411 /* ARGSUSED */
    412 int
    413 sys___settimeofday50(struct lwp *l, const struct sys___settimeofday50_args *uap,
    414     register_t *retval)
    415 {
    416 	/* {
    417 		syscallarg(const struct timeval *) tv;
    418 		syscallarg(const void *) tzp; really "const struct timezone *";
    419 	} */
    420 
    421 	return settimeofday1(SCARG(uap, tv), true, SCARG(uap, tzp), l, true);
    422 }
    423 
    424 int
    425 settimeofday1(const struct timeval *utv, bool userspace,
    426     const void *utzp, struct lwp *l, bool check_kauth)
    427 {
    428 	struct timeval atv;
    429 	struct timespec ts;
    430 	int error;
    431 
    432 	/* Verify all parameters before changing time. */
    433 
    434 	/*
    435 	 * NetBSD has no kernel notion of time zone, and only an
    436 	 * obsolete program would try to set it, so we log a warning.
    437 	 */
    438 	if (utzp)
    439 		log(LOG_WARNING, "pid %d attempted to set the "
    440 		    "(obsolete) kernel time zone\n", l->l_proc->p_pid);
    441 
    442 	if (utv == NULL)
    443 		return 0;
    444 
    445 	if (userspace) {
    446 		if ((error = copyin(utv, &atv, sizeof(atv))) != 0)
    447 			return error;
    448 		utv = &atv;
    449 	}
    450 
    451 	TIMEVAL_TO_TIMESPEC(utv, &ts);
    452 	return settime1(l->l_proc, &ts, check_kauth);
    453 }
    454 
    455 int	time_adjusted;			/* set if an adjustment is made */
    456 
    457 /* ARGSUSED */
    458 int
    459 sys___adjtime50(struct lwp *l, const struct sys___adjtime50_args *uap,
    460     register_t *retval)
    461 {
    462 	/* {
    463 		syscallarg(const struct timeval *) delta;
    464 		syscallarg(struct timeval *) olddelta;
    465 	} */
    466 	int error;
    467 	struct timeval atv, oldatv;
    468 
    469 	if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_TIME,
    470 	    KAUTH_REQ_SYSTEM_TIME_ADJTIME, NULL, NULL, NULL)) != 0)
    471 		return error;
    472 
    473 	if (SCARG(uap, delta)) {
    474 		error = copyin(SCARG(uap, delta), &atv,
    475 		    sizeof(*SCARG(uap, delta)));
    476 		if (error)
    477 			return (error);
    478 	}
    479 	adjtime1(SCARG(uap, delta) ? &atv : NULL,
    480 	    SCARG(uap, olddelta) ? &oldatv : NULL, l->l_proc);
    481 	if (SCARG(uap, olddelta))
    482 		error = copyout(&oldatv, SCARG(uap, olddelta),
    483 		    sizeof(*SCARG(uap, olddelta)));
    484 	return error;
    485 }
    486 
    487 void
    488 adjtime1(const struct timeval *delta, struct timeval *olddelta, struct proc *p)
    489 {
    490 	extern int64_t time_adjtime;  /* in kern_ntptime.c */
    491 
    492 	if (olddelta) {
    493 		mutex_spin_enter(&timecounter_lock);
    494 		olddelta->tv_sec = time_adjtime / 1000000;
    495 		olddelta->tv_usec = time_adjtime % 1000000;
    496 		if (olddelta->tv_usec < 0) {
    497 			olddelta->tv_usec += 1000000;
    498 			olddelta->tv_sec--;
    499 		}
    500 		mutex_spin_exit(&timecounter_lock);
    501 	}
    502 
    503 	if (delta) {
    504 		mutex_spin_enter(&timecounter_lock);
    505 		time_adjtime = delta->tv_sec * 1000000 + delta->tv_usec;
    506 
    507 		if (time_adjtime) {
    508 			/* We need to save the system time during shutdown */
    509 			time_adjusted |= 1;
    510 		}
    511 		mutex_spin_exit(&timecounter_lock);
    512 	}
    513 }
    514 
    515 /*
    516  * Interval timer support. Both the BSD getitimer() family and the POSIX
    517  * timer_*() family of routines are supported.
    518  *
    519  * All timers are kept in an array pointed to by p_timers, which is
    520  * allocated on demand - many processes don't use timers at all. The
    521  * first four elements in this array are reserved for the BSD timers:
    522  * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, element
    523  * 2 is ITIMER_PROF, and element 3 is ITIMER_MONOTONIC. The rest may be
    524  * allocated by the timer_create() syscall.
    525  *
    526  * Realtime timers are kept in the ptimer structure as an absolute
    527  * time; virtual time timers are kept as a linked list of deltas.
    528  * Virtual time timers are processed in the hardclock() routine of
    529  * kern_clock.c.  The real time timer is processed by a callout
    530  * routine, called from the softclock() routine.  Since a callout may
    531  * be delayed in real time due to interrupt processing in the system,
    532  * it is possible for the real time timeout routine (realtimeexpire,
    533  * given below), to be delayed in real time past when it is supposed
    534  * to occur.  It does not suffice, therefore, to reload the real timer
    535  * .it_value from the real time timers .it_interval.  Rather, we
    536  * compute the next time in absolute time the timer should go off.  */
    537 
    538 /* Allocate a POSIX realtime timer. */
    539 int
    540 sys_timer_create(struct lwp *l, const struct sys_timer_create_args *uap,
    541     register_t *retval)
    542 {
    543 	/* {
    544 		syscallarg(clockid_t) clock_id;
    545 		syscallarg(struct sigevent *) evp;
    546 		syscallarg(timer_t *) timerid;
    547 	} */
    548 
    549 	return timer_create1(SCARG(uap, timerid), SCARG(uap, clock_id),
    550 	    SCARG(uap, evp), copyin, l);
    551 }
    552 
    553 int
    554 timer_create1(timer_t *tid, clockid_t id, struct sigevent *evp,
    555     copyin_t fetch_event, struct lwp *l)
    556 {
    557 	int error;
    558 	timer_t timerid;
    559 	struct ptimers *pts;
    560 	struct ptimer *pt;
    561 	struct proc *p;
    562 
    563 	p = l->l_proc;
    564 
    565 	if ((u_int)id > CLOCK_MONOTONIC)
    566 		return (EINVAL);
    567 
    568 	if ((pts = p->p_timers) == NULL)
    569 		pts = timers_alloc(p);
    570 
    571 	pt = pool_get(&ptimer_pool, PR_WAITOK);
    572 	if (evp != NULL) {
    573 		if (((error =
    574 		    (*fetch_event)(evp, &pt->pt_ev, sizeof(pt->pt_ev))) != 0) ||
    575 		    ((pt->pt_ev.sigev_notify < SIGEV_NONE) ||
    576 			(pt->pt_ev.sigev_notify > SIGEV_SA)) ||
    577 			(pt->pt_ev.sigev_notify == SIGEV_SIGNAL &&
    578 			 (pt->pt_ev.sigev_signo <= 0 ||
    579 			  pt->pt_ev.sigev_signo >= NSIG))) {
    580 			pool_put(&ptimer_pool, pt);
    581 			return (error ? error : EINVAL);
    582 		}
    583 	}
    584 
    585 	/* Find a free timer slot, skipping those reserved for setitimer(). */
    586 	mutex_spin_enter(&timer_lock);
    587 	for (timerid = TIMER_MIN; timerid < TIMER_MAX; timerid++)
    588 		if (pts->pts_timers[timerid] == NULL)
    589 			break;
    590 	if (timerid == TIMER_MAX) {
    591 		mutex_spin_exit(&timer_lock);
    592 		pool_put(&ptimer_pool, pt);
    593 		return EAGAIN;
    594 	}
    595 	if (evp == NULL) {
    596 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
    597 		switch (id) {
    598 		case CLOCK_REALTIME:
    599 		case CLOCK_MONOTONIC:
    600 			pt->pt_ev.sigev_signo = SIGALRM;
    601 			break;
    602 		case CLOCK_VIRTUAL:
    603 			pt->pt_ev.sigev_signo = SIGVTALRM;
    604 			break;
    605 		case CLOCK_PROF:
    606 			pt->pt_ev.sigev_signo = SIGPROF;
    607 			break;
    608 		}
    609 		pt->pt_ev.sigev_value.sival_int = timerid;
    610 	}
    611 	pt->pt_info.ksi_signo = pt->pt_ev.sigev_signo;
    612 	pt->pt_info.ksi_errno = 0;
    613 	pt->pt_info.ksi_code = 0;
    614 	pt->pt_info.ksi_pid = p->p_pid;
    615 	pt->pt_info.ksi_uid = kauth_cred_getuid(l->l_cred);
    616 	pt->pt_info.ksi_value = pt->pt_ev.sigev_value;
    617 	pt->pt_type = id;
    618 	pt->pt_proc = p;
    619 	pt->pt_overruns = 0;
    620 	pt->pt_poverruns = 0;
    621 	pt->pt_entry = timerid;
    622 	pt->pt_queued = false;
    623 	timespecclear(&pt->pt_time.it_value);
    624 	if (!CLOCK_VIRTUAL_P(id))
    625 		callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
    626 	else
    627 		pt->pt_active = 0;
    628 
    629 	pts->pts_timers[timerid] = pt;
    630 	mutex_spin_exit(&timer_lock);
    631 
    632 	return copyout(&timerid, tid, sizeof(timerid));
    633 }
    634 
    635 /* Delete a POSIX realtime timer */
    636 int
    637 sys_timer_delete(struct lwp *l, const struct sys_timer_delete_args *uap,
    638     register_t *retval)
    639 {
    640 	/* {
    641 		syscallarg(timer_t) timerid;
    642 	} */
    643 	struct proc *p = l->l_proc;
    644 	timer_t timerid;
    645 	struct ptimers *pts;
    646 	struct ptimer *pt, *ptn;
    647 
    648 	timerid = SCARG(uap, timerid);
    649 	pts = p->p_timers;
    650 
    651 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    652 		return (EINVAL);
    653 
    654 	mutex_spin_enter(&timer_lock);
    655 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    656 		mutex_spin_exit(&timer_lock);
    657 		return (EINVAL);
    658 	}
    659 	if (CLOCK_VIRTUAL_P(pt->pt_type)) {
    660 		if (pt->pt_active) {
    661 			ptn = LIST_NEXT(pt, pt_list);
    662 			LIST_REMOVE(pt, pt_list);
    663 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
    664 				timespecadd(&pt->pt_time.it_value,
    665 				    &ptn->pt_time.it_value,
    666 				    &ptn->pt_time.it_value);
    667 			pt->pt_active = 0;
    668 		}
    669 	}
    670 	itimerfree(pts, timerid);
    671 
    672 	return (0);
    673 }
    674 
    675 /*
    676  * Set up the given timer. The value in pt->pt_time.it_value is taken
    677  * to be an absolute time for CLOCK_REALTIME/CLOCK_MONOTONIC timers and
    678  * a relative time for CLOCK_VIRTUAL/CLOCK_PROF timers.
    679  */
    680 void
    681 timer_settime(struct ptimer *pt)
    682 {
    683 	struct ptimer *ptn, *pptn;
    684 	struct ptlist *ptl;
    685 
    686 	KASSERT(mutex_owned(&timer_lock));
    687 
    688 	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
    689 		callout_halt(&pt->pt_ch, &timer_lock);
    690 		if (timespecisset(&pt->pt_time.it_value)) {
    691 			/*
    692 			 * Don't need to check tshzto() return value, here.
    693 			 * callout_reset() does it for us.
    694 			 */
    695 			callout_reset(&pt->pt_ch,
    696 			    pt->pt_type == CLOCK_MONOTONIC ?
    697 			    tshztoup(&pt->pt_time.it_value) :
    698 			    tshzto(&pt->pt_time.it_value),
    699 			    realtimerexpire, pt);
    700 		}
    701 	} else {
    702 		if (pt->pt_active) {
    703 			ptn = LIST_NEXT(pt, pt_list);
    704 			LIST_REMOVE(pt, pt_list);
    705 			for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list))
    706 				timespecadd(&pt->pt_time.it_value,
    707 				    &ptn->pt_time.it_value,
    708 				    &ptn->pt_time.it_value);
    709 		}
    710 		if (timespecisset(&pt->pt_time.it_value)) {
    711 			if (pt->pt_type == CLOCK_VIRTUAL)
    712 				ptl = &pt->pt_proc->p_timers->pts_virtual;
    713 			else
    714 				ptl = &pt->pt_proc->p_timers->pts_prof;
    715 
    716 			for (ptn = LIST_FIRST(ptl), pptn = NULL;
    717 			     ptn && timespeccmp(&pt->pt_time.it_value,
    718 				 &ptn->pt_time.it_value, >);
    719 			     pptn = ptn, ptn = LIST_NEXT(ptn, pt_list))
    720 				timespecsub(&pt->pt_time.it_value,
    721 				    &ptn->pt_time.it_value,
    722 				    &pt->pt_time.it_value);
    723 
    724 			if (pptn)
    725 				LIST_INSERT_AFTER(pptn, pt, pt_list);
    726 			else
    727 				LIST_INSERT_HEAD(ptl, pt, pt_list);
    728 
    729 			for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list))
    730 				timespecsub(&ptn->pt_time.it_value,
    731 				    &pt->pt_time.it_value,
    732 				    &ptn->pt_time.it_value);
    733 
    734 			pt->pt_active = 1;
    735 		} else
    736 			pt->pt_active = 0;
    737 	}
    738 }
    739 
    740 void
    741 timer_gettime(struct ptimer *pt, struct itimerspec *aits)
    742 {
    743 	struct timespec now;
    744 	struct ptimer *ptn;
    745 
    746 	KASSERT(mutex_owned(&timer_lock));
    747 
    748 	*aits = pt->pt_time;
    749 	if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
    750 		/*
    751 		 * Convert from absolute to relative time in .it_value
    752 		 * part of real time timer.  If time for real time
    753 		 * timer has passed return 0, else return difference
    754 		 * between current time and time for the timer to go
    755 		 * off.
    756 		 */
    757 		if (timespecisset(&aits->it_value)) {
    758 			if (pt->pt_type == CLOCK_REALTIME) {
    759 				getnanotime(&now);
    760 			} else { /* CLOCK_MONOTONIC */
    761 				getnanouptime(&now);
    762 			}
    763 			if (timespeccmp(&aits->it_value, &now, <))
    764 				timespecclear(&aits->it_value);
    765 			else
    766 				timespecsub(&aits->it_value, &now,
    767 				    &aits->it_value);
    768 		}
    769 	} else if (pt->pt_active) {
    770 		if (pt->pt_type == CLOCK_VIRTUAL)
    771 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual);
    772 		else
    773 			ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof);
    774 		for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list))
    775 			timespecadd(&aits->it_value,
    776 			    &ptn->pt_time.it_value, &aits->it_value);
    777 		KASSERT(ptn != NULL); /* pt should be findable on the list */
    778 	} else
    779 		timespecclear(&aits->it_value);
    780 }
    781 
    782 
    783 
    784 /* Set and arm a POSIX realtime timer */
    785 int
    786 sys___timer_settime50(struct lwp *l,
    787     const struct sys___timer_settime50_args *uap,
    788     register_t *retval)
    789 {
    790 	/* {
    791 		syscallarg(timer_t) timerid;
    792 		syscallarg(int) flags;
    793 		syscallarg(const struct itimerspec *) value;
    794 		syscallarg(struct itimerspec *) ovalue;
    795 	} */
    796 	int error;
    797 	struct itimerspec value, ovalue, *ovp = NULL;
    798 
    799 	if ((error = copyin(SCARG(uap, value), &value,
    800 	    sizeof(struct itimerspec))) != 0)
    801 		return (error);
    802 
    803 	if (SCARG(uap, ovalue))
    804 		ovp = &ovalue;
    805 
    806 	if ((error = dotimer_settime(SCARG(uap, timerid), &value, ovp,
    807 	    SCARG(uap, flags), l->l_proc)) != 0)
    808 		return error;
    809 
    810 	if (ovp)
    811 		return copyout(&ovalue, SCARG(uap, ovalue),
    812 		    sizeof(struct itimerspec));
    813 	return 0;
    814 }
    815 
    816 int
    817 dotimer_settime(int timerid, struct itimerspec *value,
    818     struct itimerspec *ovalue, int flags, struct proc *p)
    819 {
    820 	struct timespec now;
    821 	struct itimerspec val, oval;
    822 	struct ptimers *pts;
    823 	struct ptimer *pt;
    824 	int error;
    825 
    826 	pts = p->p_timers;
    827 
    828 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    829 		return EINVAL;
    830 	val = *value;
    831 	if ((error = itimespecfix(&val.it_value)) != 0 ||
    832 	    (error = itimespecfix(&val.it_interval)) != 0)
    833 		return error;
    834 
    835 	mutex_spin_enter(&timer_lock);
    836 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    837 		mutex_spin_exit(&timer_lock);
    838 		return EINVAL;
    839 	}
    840 
    841 	oval = pt->pt_time;
    842 	pt->pt_time = val;
    843 
    844 	/*
    845 	 * If we've been passed a relative time for a realtime timer,
    846 	 * convert it to absolute; if an absolute time for a virtual
    847 	 * timer, convert it to relative and make sure we don't set it
    848 	 * to zero, which would cancel the timer, or let it go
    849 	 * negative, which would confuse the comparison tests.
    850 	 */
    851 	if (timespecisset(&pt->pt_time.it_value)) {
    852 		if (!CLOCK_VIRTUAL_P(pt->pt_type)) {
    853 			if ((flags & TIMER_ABSTIME) == 0) {
    854 				if (pt->pt_type == CLOCK_REALTIME) {
    855 					getnanotime(&now);
    856 				} else { /* CLOCK_MONOTONIC */
    857 					getnanouptime(&now);
    858 				}
    859 				timespecadd(&pt->pt_time.it_value, &now,
    860 				    &pt->pt_time.it_value);
    861 			}
    862 		} else {
    863 			if ((flags & TIMER_ABSTIME) != 0) {
    864 				getnanotime(&now);
    865 				timespecsub(&pt->pt_time.it_value, &now,
    866 				    &pt->pt_time.it_value);
    867 				if (!timespecisset(&pt->pt_time.it_value) ||
    868 				    pt->pt_time.it_value.tv_sec < 0) {
    869 					pt->pt_time.it_value.tv_sec = 0;
    870 					pt->pt_time.it_value.tv_nsec = 1;
    871 				}
    872 			}
    873 		}
    874 	}
    875 
    876 	timer_settime(pt);
    877 	mutex_spin_exit(&timer_lock);
    878 
    879 	if (ovalue)
    880 		*ovalue = oval;
    881 
    882 	return (0);
    883 }
    884 
    885 /* Return the time remaining until a POSIX timer fires. */
    886 int
    887 sys___timer_gettime50(struct lwp *l,
    888     const struct sys___timer_gettime50_args *uap, register_t *retval)
    889 {
    890 	/* {
    891 		syscallarg(timer_t) timerid;
    892 		syscallarg(struct itimerspec *) value;
    893 	} */
    894 	struct itimerspec its;
    895 	int error;
    896 
    897 	if ((error = dotimer_gettime(SCARG(uap, timerid), l->l_proc,
    898 	    &its)) != 0)
    899 		return error;
    900 
    901 	return copyout(&its, SCARG(uap, value), sizeof(its));
    902 }
    903 
    904 int
    905 dotimer_gettime(int timerid, struct proc *p, struct itimerspec *its)
    906 {
    907 	struct ptimer *pt;
    908 	struct ptimers *pts;
    909 
    910 	pts = p->p_timers;
    911 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    912 		return (EINVAL);
    913 	mutex_spin_enter(&timer_lock);
    914 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    915 		mutex_spin_exit(&timer_lock);
    916 		return (EINVAL);
    917 	}
    918 	timer_gettime(pt, its);
    919 	mutex_spin_exit(&timer_lock);
    920 
    921 	return 0;
    922 }
    923 
    924 /*
    925  * Return the count of the number of times a periodic timer expired
    926  * while a notification was already pending. The counter is reset when
    927  * a timer expires and a notification can be posted.
    928  */
    929 int
    930 sys_timer_getoverrun(struct lwp *l, const struct sys_timer_getoverrun_args *uap,
    931     register_t *retval)
    932 {
    933 	/* {
    934 		syscallarg(timer_t) timerid;
    935 	} */
    936 	struct proc *p = l->l_proc;
    937 	struct ptimers *pts;
    938 	int timerid;
    939 	struct ptimer *pt;
    940 
    941 	timerid = SCARG(uap, timerid);
    942 
    943 	pts = p->p_timers;
    944 	if (pts == NULL || timerid < 2 || timerid >= TIMER_MAX)
    945 		return (EINVAL);
    946 	mutex_spin_enter(&timer_lock);
    947 	if ((pt = pts->pts_timers[timerid]) == NULL) {
    948 		mutex_spin_exit(&timer_lock);
    949 		return (EINVAL);
    950 	}
    951 	*retval = pt->pt_poverruns;
    952 	mutex_spin_exit(&timer_lock);
    953 
    954 	return (0);
    955 }
    956 
    957 /*
    958  * Real interval timer expired:
    959  * send process whose timer expired an alarm signal.
    960  * If time is not set up to reload, then just return.
    961  * Else compute next time timer should go off which is > current time.
    962  * This is where delay in processing this timeout causes multiple
    963  * SIGALRM calls to be compressed into one.
    964  */
    965 void
    966 realtimerexpire(void *arg)
    967 {
    968 	uint64_t last_val, next_val, interval, now_ns;
    969 	struct timespec now, next;
    970 	struct ptimer *pt;
    971 	int backwards;
    972 
    973 	pt = arg;
    974 
    975 	mutex_spin_enter(&timer_lock);
    976 	itimerfire(pt);
    977 
    978 	if (!timespecisset(&pt->pt_time.it_interval)) {
    979 		timespecclear(&pt->pt_time.it_value);
    980 		mutex_spin_exit(&timer_lock);
    981 		return;
    982 	}
    983 
    984 	if (pt->pt_type == CLOCK_MONOTONIC) {
    985 		getnanouptime(&now);
    986 	} else {
    987 		getnanotime(&now);
    988 	}
    989 	backwards = (timespeccmp(&pt->pt_time.it_value, &now, >));
    990 	timespecadd(&pt->pt_time.it_value, &pt->pt_time.it_interval, &next);
    991 	/* Handle the easy case of non-overflown timers first. */
    992 	if (!backwards && timespeccmp(&next, &now, >)) {
    993 		pt->pt_time.it_value = next;
    994 	} else {
    995 		now_ns = timespec2ns(&now);
    996 		last_val = timespec2ns(&pt->pt_time.it_value);
    997 		interval = timespec2ns(&pt->pt_time.it_interval);
    998 
    999 		next_val = now_ns +
   1000 		    (now_ns - last_val + interval - 1) % interval;
   1001 
   1002 		if (backwards)
   1003 			next_val += interval;
   1004 		else
   1005 			pt->pt_overruns += (now_ns - last_val) / interval;
   1006 
   1007 		pt->pt_time.it_value.tv_sec = next_val / 1000000000;
   1008 		pt->pt_time.it_value.tv_nsec = next_val % 1000000000;
   1009 	}
   1010 
   1011 	/*
   1012 	 * Don't need to check tshzto() return value, here.
   1013 	 * callout_reset() does it for us.
   1014 	 */
   1015 	callout_reset(&pt->pt_ch, pt->pt_type == CLOCK_MONOTONIC ?
   1016 	    tshztoup(&pt->pt_time.it_value) : tshzto(&pt->pt_time.it_value),
   1017 	    realtimerexpire, pt);
   1018 	mutex_spin_exit(&timer_lock);
   1019 }
   1020 
   1021 /* BSD routine to get the value of an interval timer. */
   1022 /* ARGSUSED */
   1023 int
   1024 sys___getitimer50(struct lwp *l, const struct sys___getitimer50_args *uap,
   1025     register_t *retval)
   1026 {
   1027 	/* {
   1028 		syscallarg(int) which;
   1029 		syscallarg(struct itimerval *) itv;
   1030 	} */
   1031 	struct proc *p = l->l_proc;
   1032 	struct itimerval aitv;
   1033 	int error;
   1034 
   1035 	error = dogetitimer(p, SCARG(uap, which), &aitv);
   1036 	if (error)
   1037 		return error;
   1038 	return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval)));
   1039 }
   1040 
   1041 int
   1042 dogetitimer(struct proc *p, int which, struct itimerval *itvp)
   1043 {
   1044 	struct ptimers *pts;
   1045 	struct ptimer *pt;
   1046 	struct itimerspec its;
   1047 
   1048 	if ((u_int)which > ITIMER_MONOTONIC)
   1049 		return (EINVAL);
   1050 
   1051 	mutex_spin_enter(&timer_lock);
   1052 	pts = p->p_timers;
   1053 	if (pts == NULL || (pt = pts->pts_timers[which]) == NULL) {
   1054 		timerclear(&itvp->it_value);
   1055 		timerclear(&itvp->it_interval);
   1056 	} else {
   1057 		timer_gettime(pt, &its);
   1058 		TIMESPEC_TO_TIMEVAL(&itvp->it_value, &its.it_value);
   1059 		TIMESPEC_TO_TIMEVAL(&itvp->it_interval, &its.it_interval);
   1060 	}
   1061 	mutex_spin_exit(&timer_lock);
   1062 
   1063 	return 0;
   1064 }
   1065 
   1066 /* BSD routine to set/arm an interval timer. */
   1067 /* ARGSUSED */
   1068 int
   1069 sys___setitimer50(struct lwp *l, const struct sys___setitimer50_args *uap,
   1070     register_t *retval)
   1071 {
   1072 	/* {
   1073 		syscallarg(int) which;
   1074 		syscallarg(const struct itimerval *) itv;
   1075 		syscallarg(struct itimerval *) oitv;
   1076 	} */
   1077 	struct proc *p = l->l_proc;
   1078 	int which = SCARG(uap, which);
   1079 	struct sys___getitimer50_args getargs;
   1080 	const struct itimerval *itvp;
   1081 	struct itimerval aitv;
   1082 	int error;
   1083 
   1084 	if ((u_int)which > ITIMER_MONOTONIC)
   1085 		return (EINVAL);
   1086 	itvp = SCARG(uap, itv);
   1087 	if (itvp &&
   1088 	    (error = copyin(itvp, &aitv, sizeof(struct itimerval))) != 0)
   1089 		return (error);
   1090 	if (SCARG(uap, oitv) != NULL) {
   1091 		SCARG(&getargs, which) = which;
   1092 		SCARG(&getargs, itv) = SCARG(uap, oitv);
   1093 		if ((error = sys___getitimer50(l, &getargs, retval)) != 0)
   1094 			return (error);
   1095 	}
   1096 	if (itvp == 0)
   1097 		return (0);
   1098 
   1099 	return dosetitimer(p, which, &aitv);
   1100 }
   1101 
   1102 int
   1103 dosetitimer(struct proc *p, int which, struct itimerval *itvp)
   1104 {
   1105 	struct timespec now;
   1106 	struct ptimers *pts;
   1107 	struct ptimer *pt, *spare;
   1108 
   1109 	KASSERT((u_int)which <= CLOCK_MONOTONIC);
   1110 	if (itimerfix(&itvp->it_value) || itimerfix(&itvp->it_interval))
   1111 		return (EINVAL);
   1112 
   1113 	/*
   1114 	 * Don't bother allocating data structures if the process just
   1115 	 * wants to clear the timer.
   1116 	 */
   1117 	spare = NULL;
   1118 	pts = p->p_timers;
   1119  retry:
   1120 	if (!timerisset(&itvp->it_value) && (pts == NULL ||
   1121 	    pts->pts_timers[which] == NULL))
   1122 		return (0);
   1123 	if (pts == NULL)
   1124 		pts = timers_alloc(p);
   1125 	mutex_spin_enter(&timer_lock);
   1126 	pt = pts->pts_timers[which];
   1127 	if (pt == NULL) {
   1128 		if (spare == NULL) {
   1129 			mutex_spin_exit(&timer_lock);
   1130 			spare = pool_get(&ptimer_pool, PR_WAITOK);
   1131 			goto retry;
   1132 		}
   1133 		pt = spare;
   1134 		spare = NULL;
   1135 		pt->pt_ev.sigev_notify = SIGEV_SIGNAL;
   1136 		pt->pt_ev.sigev_value.sival_int = which;
   1137 		pt->pt_overruns = 0;
   1138 		pt->pt_proc = p;
   1139 		pt->pt_type = which;
   1140 		pt->pt_entry = which;
   1141 		pt->pt_queued = false;
   1142 		if (pt->pt_type == CLOCK_REALTIME)
   1143 			callout_init(&pt->pt_ch, CALLOUT_MPSAFE);
   1144 		else
   1145 			pt->pt_active = 0;
   1146 
   1147 		switch (which) {
   1148 		case ITIMER_REAL:
   1149 		case ITIMER_MONOTONIC:
   1150 			pt->pt_ev.sigev_signo = SIGALRM;
   1151 			break;
   1152 		case ITIMER_VIRTUAL:
   1153 			pt->pt_ev.sigev_signo = SIGVTALRM;
   1154 			break;
   1155 		case ITIMER_PROF:
   1156 			pt->pt_ev.sigev_signo = SIGPROF;
   1157 			break;
   1158 		}
   1159 		pts->pts_timers[which] = pt;
   1160 	}
   1161 
   1162 	TIMEVAL_TO_TIMESPEC(&itvp->it_value, &pt->pt_time.it_value);
   1163 	TIMEVAL_TO_TIMESPEC(&itvp->it_interval, &pt->pt_time.it_interval);
   1164 
   1165 	if (timespecisset(&pt->pt_time.it_value)) {
   1166 		/* Convert to absolute time */
   1167 		/* XXX need to wrap in splclock for timecounters case? */
   1168 		switch (which) {
   1169 		case ITIMER_REAL:
   1170 			getnanotime(&now);
   1171 			timespecadd(&pt->pt_time.it_value, &now,
   1172 			    &pt->pt_time.it_value);
   1173 			break;
   1174 		case ITIMER_MONOTONIC:
   1175 			getnanouptime(&now);
   1176 			timespecadd(&pt->pt_time.it_value, &now,
   1177 			    &pt->pt_time.it_value);
   1178 			break;
   1179 		default:
   1180 			break;
   1181 		}
   1182 	}
   1183 	timer_settime(pt);
   1184 	mutex_spin_exit(&timer_lock);
   1185 	if (spare != NULL)
   1186 		pool_put(&ptimer_pool, spare);
   1187 
   1188 	return (0);
   1189 }
   1190 
   1191 /* Utility routines to manage the array of pointers to timers. */
   1192 struct ptimers *
   1193 timers_alloc(struct proc *p)
   1194 {
   1195 	struct ptimers *pts;
   1196 	int i;
   1197 
   1198 	pts = pool_get(&ptimers_pool, PR_WAITOK);
   1199 	LIST_INIT(&pts->pts_virtual);
   1200 	LIST_INIT(&pts->pts_prof);
   1201 	for (i = 0; i < TIMER_MAX; i++)
   1202 		pts->pts_timers[i] = NULL;
   1203 	mutex_spin_enter(&timer_lock);
   1204 	if (p->p_timers == NULL) {
   1205 		p->p_timers = pts;
   1206 		mutex_spin_exit(&timer_lock);
   1207 		return pts;
   1208 	}
   1209 	mutex_spin_exit(&timer_lock);
   1210 	pool_put(&ptimers_pool, pts);
   1211 	return p->p_timers;
   1212 }
   1213 
   1214 /*
   1215  * Clean up the per-process timers. If "which" is set to TIMERS_ALL,
   1216  * then clean up all timers and free all the data structures. If
   1217  * "which" is set to TIMERS_POSIX, only clean up the timers allocated
   1218  * by timer_create(), not the BSD setitimer() timers, and only free the
   1219  * structure if none of those remain.
   1220  */
   1221 void
   1222 timers_free(struct proc *p, int which)
   1223 {
   1224 	struct ptimers *pts;
   1225 	struct ptimer *ptn;
   1226 	struct timespec ts;
   1227 	int i;
   1228 
   1229 	if (p->p_timers == NULL)
   1230 		return;
   1231 
   1232 	pts = p->p_timers;
   1233 	mutex_spin_enter(&timer_lock);
   1234 	if (which == TIMERS_ALL) {
   1235 		p->p_timers = NULL;
   1236 		i = 0;
   1237 	} else {
   1238 		timespecclear(&ts);
   1239 		for (ptn = LIST_FIRST(&pts->pts_virtual);
   1240 		     ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL];
   1241 		     ptn = LIST_NEXT(ptn, pt_list)) {
   1242 			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
   1243 			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
   1244 		}
   1245 		LIST_FIRST(&pts->pts_virtual) = NULL;
   1246 		if (ptn) {
   1247 			KASSERT(ptn->pt_type == CLOCK_VIRTUAL);
   1248 			timespecadd(&ts, &ptn->pt_time.it_value,
   1249 			    &ptn->pt_time.it_value);
   1250 			LIST_INSERT_HEAD(&pts->pts_virtual, ptn, pt_list);
   1251 		}
   1252 		timespecclear(&ts);
   1253 		for (ptn = LIST_FIRST(&pts->pts_prof);
   1254 		     ptn && ptn != pts->pts_timers[ITIMER_PROF];
   1255 		     ptn = LIST_NEXT(ptn, pt_list)) {
   1256 			KASSERT(ptn->pt_type == CLOCK_PROF);
   1257 			timespecadd(&ts, &ptn->pt_time.it_value, &ts);
   1258 		}
   1259 		LIST_FIRST(&pts->pts_prof) = NULL;
   1260 		if (ptn) {
   1261 			KASSERT(ptn->pt_type == CLOCK_PROF);
   1262 			timespecadd(&ts, &ptn->pt_time.it_value,
   1263 			    &ptn->pt_time.it_value);
   1264 			LIST_INSERT_HEAD(&pts->pts_prof, ptn, pt_list);
   1265 		}
   1266 		i = TIMER_MIN;
   1267 	}
   1268 	for ( ; i < TIMER_MAX; i++) {
   1269 		if (pts->pts_timers[i] != NULL) {
   1270 			itimerfree(pts, i);
   1271 			mutex_spin_enter(&timer_lock);
   1272 		}
   1273 	}
   1274 	if (pts->pts_timers[0] == NULL && pts->pts_timers[1] == NULL &&
   1275 	    pts->pts_timers[2] == NULL && pts->pts_timers[3] == NULL) {
   1276 		p->p_timers = NULL;
   1277 		mutex_spin_exit(&timer_lock);
   1278 		pool_put(&ptimers_pool, pts);
   1279 	} else
   1280 		mutex_spin_exit(&timer_lock);
   1281 }
   1282 
   1283 static void
   1284 itimerfree(struct ptimers *pts, int index)
   1285 {
   1286 	struct ptimer *pt;
   1287 
   1288 	KASSERT(mutex_owned(&timer_lock));
   1289 
   1290 	pt = pts->pts_timers[index];
   1291 	pts->pts_timers[index] = NULL;
   1292 	if (!CLOCK_VIRTUAL_P(pt->pt_type))
   1293 		callout_halt(&pt->pt_ch, &timer_lock);
   1294 	if (pt->pt_queued)
   1295 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
   1296 	mutex_spin_exit(&timer_lock);
   1297 	if (!CLOCK_VIRTUAL_P(pt->pt_type))
   1298 		callout_destroy(&pt->pt_ch);
   1299 	pool_put(&ptimer_pool, pt);
   1300 }
   1301 
   1302 /*
   1303  * Decrement an interval timer by a specified number
   1304  * of nanoseconds, which must be less than a second,
   1305  * i.e. < 1000000000.  If the timer expires, then reload
   1306  * it.  In this case, carry over (nsec - old value) to
   1307  * reduce the value reloaded into the timer so that
   1308  * the timer does not drift.  This routine assumes
   1309  * that it is called in a context where the timers
   1310  * on which it is operating cannot change in value.
   1311  */
   1312 static int
   1313 itimerdecr(struct ptimer *pt, int nsec)
   1314 {
   1315 	struct itimerspec *itp;
   1316 
   1317 	KASSERT(mutex_owned(&timer_lock));
   1318 	KASSERT(CLOCK_VIRTUAL_P(pt->pt_type));
   1319 
   1320 	itp = &pt->pt_time;
   1321 	if (itp->it_value.tv_nsec < nsec) {
   1322 		if (itp->it_value.tv_sec == 0) {
   1323 			/* expired, and already in next interval */
   1324 			nsec -= itp->it_value.tv_nsec;
   1325 			goto expire;
   1326 		}
   1327 		itp->it_value.tv_nsec += 1000000000;
   1328 		itp->it_value.tv_sec--;
   1329 	}
   1330 	itp->it_value.tv_nsec -= nsec;
   1331 	nsec = 0;
   1332 	if (timespecisset(&itp->it_value))
   1333 		return (1);
   1334 	/* expired, exactly at end of interval */
   1335 expire:
   1336 	if (timespecisset(&itp->it_interval)) {
   1337 		itp->it_value = itp->it_interval;
   1338 		itp->it_value.tv_nsec -= nsec;
   1339 		if (itp->it_value.tv_nsec < 0) {
   1340 			itp->it_value.tv_nsec += 1000000000;
   1341 			itp->it_value.tv_sec--;
   1342 		}
   1343 		timer_settime(pt);
   1344 	} else
   1345 		itp->it_value.tv_nsec = 0;		/* sec is already 0 */
   1346 	return (0);
   1347 }
   1348 
   1349 static void
   1350 itimerfire(struct ptimer *pt)
   1351 {
   1352 
   1353 	KASSERT(mutex_owned(&timer_lock));
   1354 
   1355 	/*
   1356 	 * XXX Can overrun, but we don't do signal queueing yet, anyway.
   1357 	 * XXX Relying on the clock interrupt is stupid.
   1358 	 */
   1359 	if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL || pt->pt_queued) {
   1360 		return;
   1361 	}
   1362 	TAILQ_INSERT_TAIL(&timer_queue, pt, pt_chain);
   1363 	pt->pt_queued = true;
   1364 	softint_schedule(timer_sih);
   1365 }
   1366 
   1367 void
   1368 timer_tick(lwp_t *l, bool user)
   1369 {
   1370 	struct ptimers *pts;
   1371 	struct ptimer *pt;
   1372 	proc_t *p;
   1373 
   1374 	p = l->l_proc;
   1375 	if (p->p_timers == NULL)
   1376 		return;
   1377 
   1378 	mutex_spin_enter(&timer_lock);
   1379 	if ((pts = l->l_proc->p_timers) != NULL) {
   1380 		/*
   1381 		 * Run current process's virtual and profile time, as needed.
   1382 		 */
   1383 		if (user && (pt = LIST_FIRST(&pts->pts_virtual)) != NULL)
   1384 			if (itimerdecr(pt, tick * 1000) == 0)
   1385 				itimerfire(pt);
   1386 		if ((pt = LIST_FIRST(&pts->pts_prof)) != NULL)
   1387 			if (itimerdecr(pt, tick * 1000) == 0)
   1388 				itimerfire(pt);
   1389 	}
   1390 	mutex_spin_exit(&timer_lock);
   1391 }
   1392 
   1393 static void
   1394 timer_intr(void *cookie)
   1395 {
   1396 	ksiginfo_t ksi;
   1397 	struct ptimer *pt;
   1398 	proc_t *p;
   1399 
   1400 	mutex_enter(proc_lock);
   1401 	mutex_spin_enter(&timer_lock);
   1402 	while ((pt = TAILQ_FIRST(&timer_queue)) != NULL) {
   1403 		TAILQ_REMOVE(&timer_queue, pt, pt_chain);
   1404 		KASSERT(pt->pt_queued);
   1405 		pt->pt_queued = false;
   1406 
   1407 		if (pt->pt_proc->p_timers == NULL) {
   1408 			/* Process is dying. */
   1409 			continue;
   1410 		}
   1411 		p = pt->pt_proc;
   1412 		if (pt->pt_ev.sigev_notify != SIGEV_SIGNAL) {
   1413 			continue;
   1414 		}
   1415 		if (sigismember(&p->p_sigpend.sp_set, pt->pt_ev.sigev_signo)) {
   1416 			pt->pt_overruns++;
   1417 			continue;
   1418 		}
   1419 
   1420 		KSI_INIT(&ksi);
   1421 		ksi.ksi_signo = pt->pt_ev.sigev_signo;
   1422 		ksi.ksi_code = SI_TIMER;
   1423 		ksi.ksi_value = pt->pt_ev.sigev_value;
   1424 		pt->pt_poverruns = pt->pt_overruns;
   1425 		pt->pt_overruns = 0;
   1426 		mutex_spin_exit(&timer_lock);
   1427 		kpsignal(p, &ksi, NULL);
   1428 		mutex_spin_enter(&timer_lock);
   1429 	}
   1430 	mutex_spin_exit(&timer_lock);
   1431 	mutex_exit(proc_lock);
   1432 }
   1433 
   1434 /*
   1435  * Check if the time will wrap if set to ts.
   1436  *
   1437  * ts - timespec describing the new time
   1438  * delta - the delta between the current time and ts
   1439  */
   1440 bool
   1441 time_wraps(struct timespec *ts, struct timespec *delta)
   1442 {
   1443 
   1444 	/*
   1445 	 * Don't allow the time to be set forward so far it
   1446 	 * will wrap and become negative, thus allowing an
   1447 	 * attacker to bypass the next check below.  The
   1448 	 * cutoff is 1 year before rollover occurs, so even
   1449 	 * if the attacker uses adjtime(2) to move the time
   1450 	 * past the cutoff, it will take a very long time
   1451 	 * to get to the wrap point.
   1452 	 */
   1453 	if ((ts->tv_sec > LLONG_MAX - 365*24*60*60) ||
   1454 	    (delta->tv_sec < 0 || delta->tv_nsec < 0))
   1455 		return true;
   1456 
   1457 	return false;
   1458 }
   1459